Segmentation fault when building ISO image with nixFlakes

[rbvermaa]

Describe the bug

Segmentation fault when building ISO image with nixFlakes from release-20.09

Steps To Reproduce

To reproduce (on nixpkgs release-20.09):

nix-instantiate -A iso_minimal_new_kernel.x86_64-linux nixos/release.nix

Expected behavior

No segmentation fault

nix (Nix) 2.4pre20201102_550e11f

[roberth]

Probably #4178 which has been fixed in master. That time it wasn't reproduced by Eelco, so it's probably best if you try for yourself. You could use NixOS/nixpkgs#102932 if you think that's more convenient.

[jonringer]

For others who land on this page.

Until there's a proper fix, you may try adding GC_DONT_GC=1 to the command.

GC_DONT_GC=1 nix build ...

this seems to allow IFD to complete successfully for me

[pinpox]

For others who land on this page.

Until there's a proper fix, you may try adding GC_DONT_GC=1 to the command.

GC_DONT_GC=1 nix build ...

this seems to allow IFD to complete successfully for me

This "fixed" the problem for me aswell, but it seems more like a workaround. Any plans to get this fixed properly?

[jaen]

It has just now happened to me while building a non-ISO configuration, so it seems it's not limited to something that the ISO derivation is doing.

[tomberek]

This occurred with building an amazonImage.

• tried with flakes
• tried with legacy
  GC_DONT_GC fixed it.
  nix (Nix) 2.4pre20201205_a5d85d0

[nixos-discourse]

This issue has been mentioned on NixOS Discourse. There might be relevant details there:

https://discourse.nixos.org/t/how-can-i-debug-this-segfault-trying-to-build-my-system-with-flakes/9716/3

[stale]

I marked this as stale due to inactivity. → More info

[jonringer]

#4944 most likely fixed this, this is likely to be no longer applicable.

[ggreif]

Sorry for the linkspam above... I'll leave this comment here, as it looks like there is still some lurking bug in 2.4. There is also a fresh backtrace.

[snajpa]

yes it is, even in 2.6... I'm being bit by this or something related, flakes aren't involved in my setup, it's a NIX_PATH override with heavy overlays... prefixing nix-build with GC_DONT_GC=1 seems to avoid the segfault

[nrdxp]

hit a segfault in 2.11.1 that was resolved with the GC_DONT_GC var when building a devshell with some heavy IFD, so this bug is likely still around

[omnibs]

Same on 2.12.0, also resolved with GC_DONT_GC

[n8henrie]

Same, Segmentation fault (core dumped) building an aarch64 iso image. Works with GC_DONT_GC. Would love to learn more about how to debug these.

$ nix-shell -p nix-info --run "nix-info -m"
 - system: `"x86_64-linux"`
 - host os: `Linux 6.1.1-arch1-1, Arch Linux, noversion, rolling`
 - multi-user?: `yes`
 - sandbox: `yes`
 - version: `nix-env (Nix) 2.12.0`
 - channels(root): `"nixpkgs-22.11"`
 - channels(n8henrie): `"nixpkgs-22.11"`
 - nixpkgs: `/nix/var/nix/profiles/per-user/n8henrie/channels/nixpkgs`

[terlar]

I just hit this as well on a PR that updates my flake inputs. I could try to find out which updated input that caused the issue. This also only happens in GH Actions, running the same command locally had no issues.

[roberth]

Probably needs #7679. Is it a linux action or macos?

[terlar]

It was a Linux action

[terlar]

I have isolated the issue being triggered by nixpkgs upgrade:
terlar/nix-config@83ef162

I have not bisected anything to figure out exact at what point it broke. This breaks evaluation consistently for my GH action. Good to see a fix, will have to try it out when a new version is released. Or if anyone has a "one-liner" that could be used with install-nix-action, I guess via the install_url.

Not sure if there are any hints in this information in order to make something small and reproducible.

[nixos-discourse]

This issue has been mentioned on NixOS Discourse. There might be relevant details there:

https://discourse.nixos.org/t/rust-in-nix-discussion-thread/5092/20

[yannham]

I don't know if I'm hitting the same issue, but as @terlar, I get a segmentation fault on a nix flake check run by the CI of a PR updating flake inputs (generated by the update_flake_lock action) on a MacOS runner. The Linux runner got cancelled, so hard to say if it would be affected or not): https://github.com/tweag/nickel/actions/runs/4172595547/jobs/7223897501#step:5:23. Mentioning as a potential data point.